Biranchi Poudyal

Academic Integrity Decision Framework

A six-question framework that structures fair, consistent decisions about student generative-AI use in assessment.

Framework Ongoing · Last updated 28 July 2026

Overview

A practical decision framework for staff assessing student generative-AI use. Instead of asking the unanswerable question “did the student use AI?”, the framework structures the decision around six answerable questions that map to the concepts integrity decisions actually depend on: permission, disclosure, contribution, verification, authorship and honesty.

Problem

Integrity decisions about AI use are frequently made ad hoc: different staff weigh different factors, similar cases receive different outcomes, and students cannot predict how their behaviour will be judged. Detection tools cannot fix this because the problem is conceptual, not forensic — institutions lack a shared structure for reasoning from facts to decision.

My role

Developed the framework from my policy-analysis findings and the policy-to-decision consistency study, and tested it against the fictional case portfolio published on this site.

Research question

Can a small, fixed set of questions structure AI-related integrity decisions so that they are consistent across decision-makers, explainable to students and proportionate to the actual breach?

The framework

Every case is assessed through six questions, in order:

  1. Was AI use permitted? — What did the applicable policy and task instructions actually allow?
  2. Was it disclosed? — Did the student acknowledge the use as required?
  3. Did AI assist or replace the student’s work? — Where did the substantive intellectual contribution come from?
  4. Did the student verify the output? — Did the student check accuracy, sources and reasoning, or submit unexamined text?
  5. Was authorship preserved? — Can the student explain, defend and take responsibility for the submitted work?
  6. Was there deception? — Did the student misrepresent how the work was produced?

The pattern of answers — not any single answer — determines the outcome category and proportionate response.

Method

Conceptual synthesis from policy analysis and integrity literature, followed by iterative testing: each version of the framework was applied to the fictional case set, and questions were revised where they failed to discriminate between cases or produced unfair results.

Tools

Policy corpus from the Australian policy-analysis project, fictional case vignettes, structured worksheets for applying the framework.

Findings

The framework has not yet been formally evaluated for inter-rater consistency across independent assessors. To date its value has been demonstrated by applying it case by case to the fictional Case Portfolio published on this site — see that section for worked examples of the framework in use.

Outputs

  • The framework itself (this page and the applied examples in the Case Portfolio)
  • Staff-facing guidance based on the framework — see Technical Writing

Impact

Not yet formally measured — the framework is newly developed. The clearest evidence of its usefulness so far is qualitative: it gives staff a shared, explainable structure in place of case-by-case improvisation.

Limitations

  • The framework structures judgement; it does not replace it. Two assessors can answer the six questions differently.
  • It assumes a policy exists against which “permitted” can be answered; where policy is silent, the framework surfaces the gap but cannot fill it.
  • Tested to date on fictional cases only.

Lessons learned

  • Separating disclosure failures from authorship failures prevents the most common unfairness: treating honest, permitted use that was poorly acknowledged as if it were contract cheating.

Not yet linked to a standalone publication — the framework draws directly on the two empirical studies linked above and is documented here on this page.

Downloadable materials

A printable worksheet version of the framework is planned but not yet published here.

Try the framework

Answer the six questions below to see how the framework structures a decision. This runs entirely in your browser — nothing is submitted or stored — and is illustrative, not a formal integrity ruling.

1. Was AI use permitted for this task?
2. Was the AI use disclosed as required?
3. Did AI assist the student's work, or replace it?
4. Did the student verify the AI output?
5. Is authorship preserved — can the student explain and defend the work?
6. Was there deception about how the work was produced?

Related projects

Empirical Study Completed

Testing How University AI Governance Holds Up Under Pressure

A scenario-based method that tests whether a university's public AI-governance documents still produce an accountable decision as AI use moves from simple assistance toward more autonomous, delegated action.

Key insight — The most common weak point wasn't a missing rule — it was a named responsible role with no specific criteria, safeguards or review to back it up.

Document analysis · Scenario-based stress testing · Structured coding protocol · Independent reliability checking

Last updated 26 August 2026

Policy Analysis Completed

Do University AI Policies Actually Decide Cases?

A vignette-based test of whether the published policy environment at 20 Australian universities actually resolves common AI-assisted student behaviours, or leaves them ambiguous.

Key insight — Transparent, verified AI use was consistently the most clearly resolved by published policy — ambiguous or partially-disclosed use often wasn't resolved at all.

Vignette methodology · Structured policy classification · Dual-coder reliability testing

Last updated 26 August 2026